src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128           dummy_mask,cutoff_mask;
  77     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  78     __m128           one     = _mm_set1_ps(1.0);
  79     __m128           two     = _mm_set1_ps(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_ps(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93
  94     /* Avoid stupid compiler warnings */
  95     jnrA = jnrB = jnrC = jnrD = 0;
  96     j_coord_offsetA = 0;
  97     j_coord_offsetB = 0;
  98     j_coord_offsetC = 0;
  99     j_coord_offsetD = 0;
 100
 101     outeriter        = 0;
 102     inneriter        = 0;
 103
 104     for(iidx=0;iidx<4*DIM;iidx++)
 105     {
 106         scratch[iidx] = 0.0;
 107     }
 108
 109     /* Start outer loop over neighborlists */
 110     for(iidx=0; iidx<nri; iidx++)
 111     {
 112         /* Load shift vector for this list */
 113         i_shift_offset   = DIM*shiftidx[iidx];
 114
 115         /* Load limits for loop over neighbors */
 116         j_index_start    = jindex[iidx];
 117         j_index_end      = jindex[iidx+1];
 118
 119         /* Get outer coordinate index */
 120         inr              = iinr[iidx];
 121         i_coord_offset   = DIM*inr;
 122
 123         /* Load i particle coords and add shift vector */
 124         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 125
 126         fix0             = _mm_setzero_ps();
 127         fiy0             = _mm_setzero_ps();
 128         fiz0             = _mm_setzero_ps();
 129
 130         /* Load parameters for i particles */
 131         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 132
 133         /* Reset potential sums */
 134         velecsum         = _mm_setzero_ps();
 135
 136         /* Start inner kernel loop */
 137         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 138         {
 139
 140             /* Get j neighbor index, and coordinate index */
 141             jnrA             = jjnr[jidx];
 142             jnrB             = jjnr[jidx+1];
 143             jnrC             = jjnr[jidx+2];
 144             jnrD             = jjnr[jidx+3];
 145             j_coord_offsetA  = DIM*jnrA;
 146             j_coord_offsetB  = DIM*jnrB;
 147             j_coord_offsetC  = DIM*jnrC;
 148             j_coord_offsetD  = DIM*jnrD;
 149
 150             /* load j atom coordinates */
 151             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 152                                               x+j_coord_offsetC,x+j_coord_offsetD,
 153                                               &jx0,&jy0,&jz0);
 154
 155             /* Calculate displacement vector */
 156             dx00             = _mm_sub_ps(ix0,jx0);
 157             dy00             = _mm_sub_ps(iy0,jy0);
 158             dz00             = _mm_sub_ps(iz0,jz0);
 159
 160             /* Calculate squared distance and things based on it */
 161             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 162
 163             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 164
 165             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 166
 167             /* Load parameters for j particles */
 168             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 169                                                               charge+jnrC+0,charge+jnrD+0);
 170
 171             /**************************
 172              * CALCULATE INTERACTIONS *
 173              **************************/
 174
 175             /* Compute parameters for interactions between i and j atoms */
 176             qq00             = _mm_mul_ps(iq0,jq0);
 177
 178             /* COULOMB ELECTROSTATICS */
 179             velec            = _mm_mul_ps(qq00,rinv00);
 180             felec            = _mm_mul_ps(velec,rinvsq00);
 181
 182             /* Update potential sum for this i atom from the interaction with this j atom. */
 183             velecsum         = _mm_add_ps(velecsum,velec);
 184
 185             fscal            = felec;
 186
 187              /* Update vectorial force */
 188             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 189             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 190             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 191
 192             fjptrA             = f+j_coord_offsetA;
 193             fjptrB             = f+j_coord_offsetB;
 194             fjptrC             = f+j_coord_offsetC;
 195             fjptrD             = f+j_coord_offsetD;
 196             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 197                                                    _mm_mul_ps(dx00,fscal),
 198                                                    _mm_mul_ps(dy00,fscal),
 199                                                    _mm_mul_ps(dz00,fscal));
 200
 201             /* Inner loop uses 31 flops */
 202         }
 203
 204         if(jidx<j_index_end)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrlistA         = jjnr[jidx];
 209             jnrlistB         = jjnr[jidx+1];
 210             jnrlistC         = jjnr[jidx+2];
 211             jnrlistD         = jjnr[jidx+3];
 212             /* Sign of each element will be negative for non-real atoms.
 213              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 214              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 215              */
 216             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 217             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 218             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 219             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 220             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225
 226             /* load j atom coordinates */
 227             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                               x+j_coord_offsetC,x+j_coord_offsetD,
 229                                               &jx0,&jy0,&jz0);
 230
 231             /* Calculate displacement vector */
 232             dx00             = _mm_sub_ps(ix0,jx0);
 233             dy00             = _mm_sub_ps(iy0,jy0);
 234             dz00             = _mm_sub_ps(iz0,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 238
 239             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 240
 241             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                               charge+jnrC+0,charge+jnrD+0);
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm_mul_ps(iq0,jq0);
 253
 254             /* COULOMB ELECTROSTATICS */
 255             velec            = _mm_mul_ps(qq00,rinv00);
 256             felec            = _mm_mul_ps(velec,rinvsq00);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velec            = _mm_andnot_ps(dummy_mask,velec);
 260             velecsum         = _mm_add_ps(velecsum,velec);
 261
 262             fscal            = felec;
 263
 264             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 265
 266              /* Update vectorial force */
 267             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 268             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 269             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 270
 271             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 272             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 273             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 274             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 275             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 276                                                    _mm_mul_ps(dx00,fscal),
 277                                                    _mm_mul_ps(dy00,fscal),
 278                                                    _mm_mul_ps(dz00,fscal));
 279
 280             /* Inner loop uses 31 flops */
 281         }
 282
 283         /* End of innermost loop */
 284
 285         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 286                                               f+i_coord_offset,fshift+i_shift_offset);
 287
 288         ggid                        = gid[iidx];
 289         /* Update potential energies */
 290         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 291
 292         /* Increment number of inner iterations */
 293         inneriter                  += j_index_end - j_index_start;
 294
 295         /* Outer loop uses 8 flops */
 296     }
 297
 298     /* Increment number of outer iterations */
 299     outeriter        += nri;
 300
 301     /* Update outer/inner flops */
 302
 303     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 304 }
 305 /*
 306  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 307  * Electrostatics interaction: Coulomb
 308  * VdW interaction:            None
 309  * Geometry:                   Particle-Particle
 310  * Calculate force/pot:        Force
 311  */
 312 void
 313 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 314                     (t_nblist * gmx_restrict                nlist,
 315                      rvec * gmx_restrict                    xx,
 316                      rvec * gmx_restrict                    ff,
 317                      t_forcerec * gmx_restrict              fr,
 318                      t_mdatoms * gmx_restrict               mdatoms,
 319                      nb_kernel_data_t * gmx_restrict        kernel_data,
 320                      t_nrnb * gmx_restrict                  nrnb)
 321 {
 322     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 323      * just 0 for non-waters.
 324      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 325      * jnr indices corresponding to data put in the four positions in the SIMD register.
 326      */
 327     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 328     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 329     int              jnrA,jnrB,jnrC,jnrD;
 330     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 331     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 332     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 333     real             rcutoff_scalar;
 334     real             *shiftvec,*fshift,*x,*f;
 335     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 336     real             scratch[4*DIM];
 337     __m128           fscal,rcutoff,rcutoff2,jidxall;
 338     int              vdwioffset0;
 339     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 340     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 341     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 342     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 343     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 344     real             *charge;
 345     __m128           dummy_mask,cutoff_mask;
 346     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 347     __m128           one     = _mm_set1_ps(1.0);
 348     __m128           two     = _mm_set1_ps(2.0);
 349     x                = xx[0];
 350     f                = ff[0];
 351
 352     nri              = nlist->nri;
 353     iinr             = nlist->iinr;
 354     jindex           = nlist->jindex;
 355     jjnr             = nlist->jjnr;
 356     shiftidx         = nlist->shift;
 357     gid              = nlist->gid;
 358     shiftvec         = fr->shift_vec[0];
 359     fshift           = fr->fshift[0];
 360     facel            = _mm_set1_ps(fr->epsfac);
 361     charge           = mdatoms->chargeA;
 362
 363     /* Avoid stupid compiler warnings */
 364     jnrA = jnrB = jnrC = jnrD = 0;
 365     j_coord_offsetA = 0;
 366     j_coord_offsetB = 0;
 367     j_coord_offsetC = 0;
 368     j_coord_offsetD = 0;
 369
 370     outeriter        = 0;
 371     inneriter        = 0;
 372
 373     for(iidx=0;iidx<4*DIM;iidx++)
 374     {
 375         scratch[iidx] = 0.0;
 376     }
 377
 378     /* Start outer loop over neighborlists */
 379     for(iidx=0; iidx<nri; iidx++)
 380     {
 381         /* Load shift vector for this list */
 382         i_shift_offset   = DIM*shiftidx[iidx];
 383
 384         /* Load limits for loop over neighbors */
 385         j_index_start    = jindex[iidx];
 386         j_index_end      = jindex[iidx+1];
 387
 388         /* Get outer coordinate index */
 389         inr              = iinr[iidx];
 390         i_coord_offset   = DIM*inr;
 391
 392         /* Load i particle coords and add shift vector */
 393         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 394
 395         fix0             = _mm_setzero_ps();
 396         fiy0             = _mm_setzero_ps();
 397         fiz0             = _mm_setzero_ps();
 398
 399         /* Load parameters for i particles */
 400         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 401
 402         /* Start inner kernel loop */
 403         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrA             = jjnr[jidx];
 408             jnrB             = jjnr[jidx+1];
 409             jnrC             = jjnr[jidx+2];
 410             jnrD             = jjnr[jidx+3];
 411             j_coord_offsetA  = DIM*jnrA;
 412             j_coord_offsetB  = DIM*jnrB;
 413             j_coord_offsetC  = DIM*jnrC;
 414             j_coord_offsetD  = DIM*jnrD;
 415
 416             /* load j atom coordinates */
 417             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                               x+j_coord_offsetC,x+j_coord_offsetD,
 419                                               &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm_sub_ps(ix0,jx0);
 423             dy00             = _mm_sub_ps(iy0,jy0);
 424             dz00             = _mm_sub_ps(iz0,jz0);
 425
 426             /* Calculate squared distance and things based on it */
 427             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 428
 429             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 430
 431             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 432
 433             /* Load parameters for j particles */
 434             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 435                                                               charge+jnrC+0,charge+jnrD+0);
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             /* Compute parameters for interactions between i and j atoms */
 442             qq00             = _mm_mul_ps(iq0,jq0);
 443
 444             /* COULOMB ELECTROSTATICS */
 445             velec            = _mm_mul_ps(qq00,rinv00);
 446             felec            = _mm_mul_ps(velec,rinvsq00);
 447
 448             fscal            = felec;
 449
 450              /* Update vectorial force */
 451             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 452             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 453             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 454
 455             fjptrA             = f+j_coord_offsetA;
 456             fjptrB             = f+j_coord_offsetB;
 457             fjptrC             = f+j_coord_offsetC;
 458             fjptrD             = f+j_coord_offsetD;
 459             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 460                                                    _mm_mul_ps(dx00,fscal),
 461                                                    _mm_mul_ps(dy00,fscal),
 462                                                    _mm_mul_ps(dz00,fscal));
 463
 464             /* Inner loop uses 30 flops */
 465         }
 466
 467         if(jidx<j_index_end)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrlistA         = jjnr[jidx];
 472             jnrlistB         = jjnr[jidx+1];
 473             jnrlistC         = jjnr[jidx+2];
 474             jnrlistD         = jjnr[jidx+3];
 475             /* Sign of each element will be negative for non-real atoms.
 476              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 477              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 478              */
 479             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 480             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 481             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 482             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 483             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 484             j_coord_offsetA  = DIM*jnrA;
 485             j_coord_offsetB  = DIM*jnrB;
 486             j_coord_offsetC  = DIM*jnrC;
 487             j_coord_offsetD  = DIM*jnrD;
 488
 489             /* load j atom coordinates */
 490             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 491                                               x+j_coord_offsetC,x+j_coord_offsetD,
 492                                               &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm_sub_ps(ix0,jx0);
 496             dy00             = _mm_sub_ps(iy0,jy0);
 497             dz00             = _mm_sub_ps(iz0,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 501
 502             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 503
 504             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 505
 506             /* Load parameters for j particles */
 507             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 508                                                               charge+jnrC+0,charge+jnrD+0);
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm_mul_ps(iq0,jq0);
 516
 517             /* COULOMB ELECTROSTATICS */
 518             velec            = _mm_mul_ps(qq00,rinv00);
 519             felec            = _mm_mul_ps(velec,rinvsq00);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 524
 525              /* Update vectorial force */
 526             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 527             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 528             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 529
 530             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 531             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 532             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 533             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 534             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 535                                                    _mm_mul_ps(dx00,fscal),
 536                                                    _mm_mul_ps(dy00,fscal),
 537                                                    _mm_mul_ps(dz00,fscal));
 538
 539             /* Inner loop uses 30 flops */
 540         }
 541
 542         /* End of innermost loop */
 543
 544         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 545                                               f+i_coord_offset,fshift+i_shift_offset);
 546
 547         /* Increment number of inner iterations */
 548         inneriter                  += j_index_end - j_index_start;
 549
 550         /* Outer loop uses 7 flops */
 551     }
 552
 553     /* Increment number of outer iterations */
 554     outeriter        += nri;
 555
 556     /* Update outer/inner flops */
 557
 558     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 559 }